Research Experience

Have the courage to follow your heart and intuition. ​——Steve Jobs

#Photographed at East Lake, Wuhan

Learning Geospatial Region Embedding with Heterogeneous Graph  [paper][code]

Learning geospatial embeddings is essential for urban and environmental analysis. We propose GeoHG, a graph structure that learns comprehensive region embeddings by integrating satellite image features with POIs and socio-environmental data. Despite data scarcity, GeoHG demonstrates superior performance in prediction tasks and generalizes well across different regions.

Deep Learning for Cross-Domain Data Fusion in Urban Computing: Taxonomy, Advances, and Outlook  [paper] [ code ]

Urban Computing is becoming crucial for sustainable development, leveraging cross-domain data fusion from various sources and modalities. Deep learning methods are increasingly used for data fusion in smart cities. We present the first systematic review of deep learning-based data fusion methods in urban computing. We explore data perspectives, classify methodologies into four categories, and categorize urban applications into seven types. Our review emphasizes the synergy between deep learning and urban computing, discussing the potential impact of Large Language Models (LLMs) and suggesting future research directions. This taxonomy and analysis aim to enhance the research community.

Model for Prediction of Rock Joint Roughness and Based Convolutional Neural Network [paper]

The presence of structural planes (rock joints) in rock will constrain its mechanical performance, and the roughness of these planes controls this effect. Traditional approaches are often limited to two-dimensional structures and empirical formulas. We will attempt to use a non-contact method to scan the three-dimensional structure of the surface and use deep learning to automatically extract its features. We constructed a non-contact workflow for quantifying the roughness of structural planes and could predict the overall shear strength of the rock with the accuracy of JRC prediction over 75%. This research is supervised by  Dr. Qi Zhao . More experiments are needed, looking for your cooperation. 

Smart Classifier for Worker's Working Posture for Rapid Entire Body Assessment (REBA)

The work posture of workers on construction sites may damage their physical health, and prolonged poor posture can have irreversible effects on their bones and muscles. Some scholars have proposed using REBA to quantify this damage and ensure the health of workers through reasonable management. Traditional evaluation methods usually rely on supervisors to visually assess workers, which is inefficient and makes it difficult to promote the use of REBA in practical engineering.   This study attempts to build a posture recognition model based on workers' mobile phones and built-in sensor data, using a special framework of CNN+LSTM through a time-series neural network. Then predict REBA values through posture information. This research is supervised by  Dr. Yantao Yu .

Automated Solutions for Large-scale Waterproofing Construction; JBOT Limited (CR NO. 3223788 HK)

JBOT (HK) LIMITED , (CR NO. 3223788), led by Yu Xueqing  (CEO) and Zou Xingchen (Co-founder), has developed a robot that combines computer vision technology with on-site construction. This robot is capable of handling most job environments. With its clever mechanical structure and sensitive vision alignment system, combined with pre-mixed combustion technology, it can ensure a construction completion rate of over 95% while saving materials. Based on this robot, we have designed a complete intelligent construction system that incorporates Internet of Things (IoT) technology, achieving a perfect integration of waterproof construction process intelligence and automation. 


Extreme Temperature Distribution in Concrete Bridges Under Climate Change [paper]
Temperature effects are critical in structural design, as overall high temperatures and uneven temperature distribution of materials can cause significant thermal loads inside structures. Many existing structural design manuals in Hong Kong and other regions consider temperature loads, but they often overlook the impact of global warming. Due to the emissions of carbon dioxide, global climate will continue to warm, leading to increasingly extreme high temperatures worldwide that often exceed the temperatures considered in design manuals. This may have fatal consequences for existing buildings designed based on these manuals. Large bridges are the most temperature-sensitive type of structure. This study will use historical observation data of the Shenzhen Bay Bridge and meteorological data provided by the Hong Kong Observatory to predict climate change in local areas of Hong Kong. Through Abaqus, the study will establish and calibrate a finite element thermodynamic model of the bridge, predict the thermal response of the bridge under extreme climate after warming, and provide structural reinforcement recommendations based on the results. This research is primarily included in my graduation thesis and has received support from the Architectural Services Department of Hong Kong and was under the guidance from  Prof. F.T.K. Au.


( To be updated   🙂 )

Website Templates

created with

Website Builder Software .